Simulate auto-correlations and improve calibration solutions

[cjordan]

This work is started in the autos branch.

One big thing that needs to change is the modelling code's assumption that all incoming baselines are cross-correlation baselines. A couple of formulas are used to work out tile indices from a CC baseline index, so this will need to be adjusted depending on whether autos are also in there. The rest of the modelling code should be OK; a simulated auto vis should just be B S B^H, where B is its beam response and S is the flux density Jones matrix.

However the rest of the code needs to know if autos are being used too. The input data might not have any autos at all (so don't bother simulating autos in this case), and/or the user says to ignore autos. Juggling these flags is a bit annoying but I've started this work; hopefully it makes sense.

When autos can be simulated, Wenyang's approach to improving calibration solutions can be implemented. I've also started this work, but I think it's incorrect.